sftp/scp performance testing

[Dan Kaminsky]

> I once did something similar to test the performance of an ssh-based vpn
> (but I used ttcp). And got quite surprised when I got a throughput over
> 100Mbit/s when the vpn-nodes were connected with a 10Mbit/s network. The
> answer however was simple, they had compression enabled:-) From that
> point on I always use random-data to do performance testing.

Yes, compression will break /dev/zero mightily :-)  But /dev/urandom just
isn't as fast, at least not on all platforms.

Piping nonrandom data into SSH completely randomizes that data from the
perspective of any routes in the middle, due to the equivalence of encrypted
data and pure entropy.

--Dan